AIHugging Faceabout 2 hours ago
Hugging Face Releases Olmo-core 3 Training Stack
7 min read
Hugging Face released Olmo-core 3, a redesigned open mixture-of-experts training framework for large language models. It scales to 1.2 trillion parameters across 512 Nvidia B300 GPUs, and a 47-billion-parameter MoE hit 52,000 tokens per second per GPU, about 2.7 times its earlier implementation. The stack underpins the next-generation Olmo model.
Level
Hype check
Tap to vote and see what everyone thinks.
